9

I applied a new method to a data set, which had already been analyzed in a research paper. My method is generally considered better, but it only slightly modifies the results and does not affect the conclusion of the paper aside from improving the accuracy of results by a small amount. The paper is about a classification problem.

Is it possible to write a research paper about this?

  • 2
    Are you asking if this is scientifically acceptable? This doesn't seem to be exactly about statistics; it may be a better fit on academia.SE. – gung Sep 20 '14 at 14:10
  • 1
    As written, this question is incredibly vague. In general this can be and has been done. Are you concerned with the statistical validity of your new technique, or the legitimacy of re-using the prior data set, or both? – user479 Sep 20 '14 at 15:13
  • @Wrzlprmft very loosely, a classification problem is a computational/statistical problem where you try to find a mathematical function that will successfully distinguish between two types of objects based on their numerical properties. en.wikipedia.org/wiki/Statistical_classification – Bitwise Sep 20 '14 at 18:22
  • @Bitwise: I see; I thought this was somehow referring to the issue itself and not the content of the paper. – Wrzlprmft Sep 20 '14 at 19:02
  • @Wrzlprmft I think the scope of the 'punishability' tag should be limited to questions of novelty, significance, and scope. Authorship, ethics, and copyright issues are really a completely different category. – ff524 Sep 21 '14 at 2:09
8

I'd say it is difficult to answer this question in general. You should ask yourself if you are contributing something significantly novel to the academic discussion. For example, if you reanalyse the dataset and get completely different or new results, then it is probably worth publishing. On the other hand, if you only slightly modify the statistical method and arrive at the same results, it is probably not worth it. Before beginning the analysis, you could ask yourself if this dataset is particularly big or of good quality or if there are other, better datasets out there (in which case a reanalysis of this particular one might be unproductive).

To give an example, in the UK there have been many papers on the effectiveness of breast cancer screening that essentially analysed the same dataset using different methods. These methods were fairly complex and the various approaches have significantly advances our understanding of the advantages and possible detriments of breast cancer screening. In this case, publishing multiple papers was justified.

  • If the new method is significantly more efficient (e.g. able to obtain the solution in minutes rather than hours or even days) then it would be worth publishing even if the conclusions are exactly the same. Indeed, in such scenario you may be looking for finding the same conclusion, just more efficiently. – ddiez Sep 21 '14 at 12:26
8

In the bioinformatics field, for example, it is perfectly acceptable to use the same dataset for a publication demonstrating the analysis using a different method. There are (at least) two possibilities:

  • Both the dataset and the method are not novel, but applying the new method on the dataset results in novel insight about the studied system. An example could be two recent papers in PNAS. In one a dataset and some statistical analysis approach was used to demonstrate divergent inflammatory responses in mouse and human. In the second paper other group used the same dataset with some other (not novel) methods to demonstrate the contrary.

  • The dataset is not new but the method is novel. This case is very common in Bioinformatics and aims to use a dataset for which the outcome is relatively well known. It could also be used to demonstrate how the new method leads to novel insight.

I would stress that in the first case the new insight aspect has more weight as neither the dataset or the method are novel. In the second case this is not so important, as far as the new method results in some other advantage.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy